HTML
--> --> -->The atmospheric geopotential height over the polar cap can be used to identify changes in circulation, and is strongly related to the mean temperature below a particular level. Changes in temperatures at lower levels can impact the atmosphere above through this relationship. The work by (Sun et al., 2015) showed that changes in sea ice can impact the mid-to-upper tropospheric and lower stratospheric circulation in an idealized model. This is supported by the results of (Peings and Magnusdottir, 2014), (Kim et al., 2014) and (Nakamura et al., 2016), among others, who identified a connection with the stratosphere. However, such studies were largely concerned with the response to low sea ice and did not explicitly consider the causes of low sea ice in the first instance. In the present study, we suggest that the atmospheric conditions that precede low sea ice can also weaken the polar stratospheric vortex directly.
There is an established relationship between mid-tropospheric eddy meridional heat flux, which is the vertical component of Eliassen-Palm wave activity flux (Edmon et al., 1980), and stratospheric circulation (e.g., Newman and Nash, 2000; Sjoberg and Birner, 2012). Enhanced heat flux, or upward wave activity relative to climatology, over a period of a few months has been shown to be related to a weakened stratospheric polar vortex. Thereafter, changes in polar stratospheric circulation have been shown to enhance long time scale predictability in the troposphere (Baldwin and Dunkerton, 2001; Christiansen, 2001; Baldwin et al., 2003). As enhanced tropospheric meridional heat flux affects the stratosphere, and is also associated with midlatitude circulation anomalies that may affect sea ice, we speculate that midlatitude circulation anomalies associated with positive eddy heat flux anomalies can affect both Arctic sea ice and the stratospheric circulation directly, and then provide evidence for this speculation. This complicates the assessment of causality of sea-ice-stratosphere linkages, as both may be responding to eddy heat flux anomalies rather than the sea ice driving the stratosphere directly.
A hypothesized impact of Arctic sea ice loss is the "warm Arctic, cold continent" pattern in surface temperatures. Work by (Mori et al., 2014) suggested that more frequent Eurasian blocking due to sea ice loss forces cold-air advection into the region and thus cooler Eurasian winters. In (Petoukhov and Semenov, 2010) a similar process was discussed, though the resulting pattern was found to be nonlinearly dependent on the degree of sea ice loss. The work of (McCusker et al., 2016) and (Sun et al., 2016), however, provided modelling evidence that while a warm Arctic is driven by sea ice loss, the cold continental temperature pattern may not be. (Sorokina et al., 2016) found a robust relationship between turbulent heat flux and Barents-Kara Sea ice using reanalysis data, though the link to cold continental temperatures was not apparent. As the "warm Arctic, cold continent" is a pattern that can be driven purely by internal variability, we attempt, by means of lead-lag regressions, to elucidate its temporal evolution and infer the directionality of the relationship between this pattern and Arctic sea ice.
The atmospheric response to sea ice loss is likely to be sensitive to the geographical location of the ice anomalies. The results of (Petoukhov and Semenov, 2010), (Sun et al., 2015), (Koenigk et al., 2016), (Pedersen et al., 2016), (Screen, 2017b) and others suggest that different regions of ice loss have different response patterns. This is possibly related to the interference with the climatological mean planetary wave (Martius et al., 2009; Garfinkel et al., 2010; Smith et al., 2011), whereby constructive (destructive) interference between the forced and climatological planetary waves acts to enhance (suppress) vertical wave propagation. For this reason, the present study examines relationships with pan-Arctic anomalies as well as with regional sea ice anomalies.
This paper seeks to further our understanding of the precursors of, and response to, Arctic sea ice loss, presenting evidence from CMIP5 climate models. The CMIP5 models have been a relatively underused resource in this regard, with the notable exception of (Boland et al., 2017). In contrast to (Boland et al., 2017), who examined historical and future scenarios, we focus on pre-industrial control simulations to examine the internal variability in the absence of forced trends. We are especially motivated to better understand the nature of the coupled two-way relationship between Arctic sea ice and the stratospheric polar vortex, and additionally between Arctic sea ice and cold winter temperatures over Eurasia, as present within the selected CMIP5 models.
Time series of isobaric geopotential height and sea ice area (sea ice concentration multiplied by grid cell area) over the polar cap (66°-90°N), and zonal-mean meridional eddy heat flux (\(\overline{v'T'}\)) over 45°-65°N are calculated. This latitudinal band for the averaging of heat flux is different to that used in previous studies (e.g., Newman and Nash, 2000; Sjoberg and Birner, 2012) in order to separate the Arctic and midlatitudes, but this choice does not affect the results qualitatively. The time series of each variable is then used to calculate standardized climatological anomalies, \(x'=(\bar{x}-x)/s_x\), where \(\bar{x}\) and sx are the long-term monthly mean and standard deviation for the nearest 30 years to the modelled monthly variable x.
The standardized sea ice area anomalies are regressed against the diagnostic variable anomalies at leads and lags of up to 14 months. To generate seasonal regressions, sea ice is masked such that only anomalies from each individual season are regressed against the atmospheric variables of all seasons. This means a lag of -3 in the December-January-February (DJF) mean is a mean of the regression of all December sea ice with September diagnostic variables, January sea ice with October diagnostic variables, and February sea ice with November diagnostic variables. In all plots, we show the negative regression slope, as this enables the association between the atmosphere and low sea ice conditions to be demonstrated. Statistical significance is calculated using Fisher's method (Kost and McDermott, 2002).
3.1. Polar cap height
We begin by looking at the linear regression between sea ice area and polar cap height at extended lead and lag times (Fig. 1) for the set of all CMIP5 models with data available. There are statistically significant regressions at both positive and negative lags, implying both atmospheric precursors of, and responses to, low sea ice. Positive (anticyclonic) anomalies are the dominant signal through most of the atmosphere, with significant anomalies at both positive and negative lag times. This indicates that low sea ice is preceded by, and followed by, tropospheric high geopotential height anomalies in the Arctic region. In general, the regressions are stronger when the atmosphere leads sea ice, which suggests that sea ice, at least initially, is not forcing the changes in the polar mid-to-upper troposphere and lower stratosphere. This is especially the case for low summer and low autumn sea ice, suggesting sea ice in these seasons is particularly sensitive to the atmospheric conditions in preceding months. However, there are statistically significant positive anomalies in polar cap height following low sea ice in all seasons, but especially in winter and spring, which suggests a weakening of the polar stratospheric vortex following low sea ice.Figure 2 is constructed in a similar manner to Fig. 1, but for the subset of independent models described in Table 1. There are some small but notable differences in the magnitude of the regressions (between the subset and full set), but the largest differences are in the areas of model agreement and statistical significance. Figure 3, showing the results from a subset of high-top (greater than 0.01 hPa) models, is similar to the previous two figures, with some key exceptions. The magnitudes of the regression slopes are higher, and all seasons show a statistically significant but weak negative anomaly in the early spring stratosphere. The latter may indicate a more persistent polar stratospheric vortex in spring, relative to the climatological mean transition to anticyclonic summer circulation. This could be a delay in the final stratospheric warming, the transition between winter (cyclonic) and summer (anticyclonic) stratospheric circulations, typically occurring in April. In general, the qualitative differences in the regressions are small (comparing the high-top subset and the full set), but the high-top subset has a smaller area of statistical significance and robustness compared to the full set. In the following figures, we show results only from the models of the first subset of independent models, as the differences between the two subsets and the full set are small. It should also be noted that the maximum regression slopes, as well as correlation coefficients, are small (maximums of 0.3), despite the relationship being robust across models and statistically significant. This is to be expected, as multiple factors influence atmospheric circulation in addition to sea ice.
Figure1. Linear regression of standardized polar cap sea ice area anomalies against standardized polar cap geopotential height anomalies for each season for the full set of models. The regressions have been multiplied by minus one to show the patterns associated with low sea ice. Hatching covers areas not statistically significant at the 99% confidence level, while dots cover areas where fewer than 75% of models agree with the sign of the regression slope. Negative lags indicate atmosphere leading sea ice. The shading is unitless (standardized regression coefficient).
Figure2. As in Fig. 1, but for the subset of eight selected models shown in Table 1.
Figure3. As in Fig. 1, but for the subset of six selected high-top models shown in Table 1.2
3.2. Eddy heat flux
We now turn our attention to the midlatitude tropospheric meridional eddy heat flux (hereafter, "heat flux"), which is known to drive stratospheric variability and is the vertical component of the Eliassen-Palm flux. In all seasons, a statistically significant heat flux is found to precede anomalously low polar cap sea ice (Fig. 4). Enhanced heat flux is apparent in the lower troposphere for up to 6 months prior to low sea ice in winter and spring, and 12 months prior to low sea ice in summer and autumn. This strongly suggests that enhanced poleward heat flux contributes to the low sea ice anomalies. There is little evidence for the opposite——sea ice causing a change in the heat flux——with mostly insignificant regressions at positive lag times (i.e., following anomalously low ice). A positive heat flux is known to contribute to stratospheric polar vortex weakening. The heat flux anomalies preceding low sea ice are one likely cause of the enhanced polar cap height that also precedes low sea ice. Therefore, it is probable that the sea ice and polar cap height are both responding to this enhanced midlatitude heat flux——similar to the results of (Perlwitz et al., 2015) and (Screen et al., 2012) with respect to Arctic warming being driven by heat transport into the Arctic from lower latitudes.
Figure4. As in Fig. 2 but for midlatitude meridional eddy heat flux standardized anomalies. The shading is unitless (standardized regression coefficient).2
3.3. Surface temperature
In previous work, low sea ice (and in some cases a weakened stratospheric polar vortex) has been proposed to cause the "warm Arctic, cold continent" winter temperature anomaly pattern. It has been argued that low Arctic sea ice causes warmer Arctic surface temperatures but cooler conditions over Eurasia and North America (Honda et al., 2009; Petoukhov and Semenov, 2010; Cohen et al., 2013; Mori et al., 2014; Kug et al., 2015). Figure 5 shows the lead-lag relationship between winter sea ice and Northern Hemisphere surface temperature. The CMIP5 models reproduce the "warm Arctic, cold continent" anomaly pattern at zero lag, with significant cold winter temperature anomalies over Eurasia correlated with low winter sea ice. This temperature anomaly pattern is also seen at a lag of -1 month and, to a lesser extent, at a lag of -2 months. This implies that both the Arctic warming and Eurasian cooling precede low winter sea ice.
Figure5. Linear regression of winter (DJF) polar cap sea ice area standardized anomalies against standardized surface temperature anomalies between lag -1 months to lead +4 months. The regressions have been multiplied by minus one to show the patterns associated with low sea ice. Blue, dashed contours are cold anomalies; red, solid contours are warm anomalies. The shading is unitless (standardized regression coefficient).The warm anomaly in the Arctic is maximized over the Barents-Kara Sea and is present for at least two months before low winter sea ice. The progression of anomalously warm Arctic temperatures supports the results presented in the previous section, where warmer midlatitude air is transported to the Arctic, thereby reducing sea ice. The warm anomaly persists over the Barents-Kara Sea at lags of up to 3 months, likely in response to the low sea ice. The cool continental anomaly, however, is only present in the months before low sea ice, and not after. This implies the Eurasian cooling is not a response to low sea ice, but instead is driven by atmospheric circulation changes that precede and contribute to low sea ice. Of note is that we also find no evidence for Eurasian winter cooling following low sea ice in other seasons. More specifically, we find no evidence for Eurasian winter cooling following low autumn sea ice, as suggested by others (e.g., Francis et al., 2009; Hopsch et al., 2012; Jaiser et al., 2012).
2
3.4. Sea level pressure
To further examine the atmospheric circulation changes linked to the Eurasian cooling, we carry out the same analysis again but with sea level pressure. In the CMIP5 models, the Eurasian cooling is dynamically related to a strengthened Siberian high, consistent with previous studies (Mori et al., 2014; Sun et al., 2016). A high sea level pressure anomaly is found simultaneously with, and for two months prior to, low winter sea ice, which can be seen in Fig. 6. The strengthened Siberian high appears part of a larger-scale pattern of circulation anomalies, including a positive North Atlantic Oscillation (NAO)-type pattern in the North Atlantic and raised pressure in the North Pacific. The surface circulation anomalies are much weaker at positive lags, with the most notable feature being a negative NAO pattern at lags of 1 and 2 months. There is no evidence of a strengthened Siberian high following low sea ice, which helps explain the lack of Eurasian cooling following low winter sea ice.
Figure6. As in Fig. 5 but for standardized mean sea level pressure anomalies. Red, solid contours are high pressure anomalies; blue, dashed contours are low pressure anomalies. The shading is unitless (standardized regression coefficient).Several studies have examined the Siberian winter cooling trend, some of which have found that sea ice loss is a precursor to cold continental temperatures (Petoukhov and Semenov, 2010; Mori et al., 2014). Others, meanwhile, have found that sea ice does not drive the cold continental temperatures, but does force a warming Arctic (McCusker et al., 2016; Sorokina et al., 2016; Sun et al., 2016). Our study falls into the latter category insofar as that, while there is evidence for sea ice loss as a precursor to warmer Arctic surface temperatures, the same cannot be said for cold continental temperatures. Thus far, the causes of the "warm Arctic, cold continent" pattern remain uncertain, as discussed in (Screen, 2017a).
Figure7. Geographic regions used for spatial averaging of atmospheric and sea ice variables. Grey is the polar cap; Barents-Kara Sea in blue; Bering Sea in orange; Sea of Okhotsk in red; Greenland Sea in green.2
3.5. Regional sea ice anomalies
As mentioned earlier, low ice in specific regions of the Arctic can impact the atmosphere in different ways. To examine these relationships, Arctic sea ice is partitioned into the marginal seas shown in Fig. 7, based on those previously used in (Screen, 2017b). Figures 8 and 9 show regressions of sea ice, averaged over the four selected polar seas, against the polar cap geopotential height and eddy heat flux, respectively. The regressions of Barents-Kara Sea winter sea ice with polar cap height (Fig. 8a) are similar to those previously shown for the pan-Arctic ice area, with positive polar cap height (Fig. 8a) and eddy heat flux (Fig. 9a) anomalies preceding low ice by 2-3 months, and positive polar cap height anomalies following low sea ice. However, in comparison to the regressions with the pan-Arctic sea ice area, the regressions against Barents-Kara Sea ice are weaker at negative lags and strong at positive lags. Broadly similar lead and lag regressions are found for low winter Greenland Sea ice (Figs. 8b and 9b). There are significant (mainly tropospheric) positive polar cap height and eddy heat flux anomalies preceding, and coincident with, low winter Bering Sea ice (Figs. 8c and 9c). The Sea of Okhotsk has a noticeably distinct pattern from the other seas, with a large negative polar cap height anomaly (Fig. 8d) and negative heat flux (Fig. 9d) in the 2-5 months prior to low ice. This indicates reduced vertical wave activity propagation into the stratosphere and a stronger polar vortex.
Figure8. As in Fig. 2 but for winter (DJF) sea ice standardized anomalies in the (a) Barents-Kara Sea, (b) Bering Sea, (c) Greenland Sea and (d) Sea of Okhotsk.
Figure9. As in Fig. 8 but for midlatitude meridional eddy heat flux standardized anomalies. The shading is unitless (standardized regression coefficient).
Figure10. As in Fig. 2, but for RCP8.5 simulations.As year-on-year variations in sea ice during the pre-industrial control simulations may be of different magnitude and spatial pattern to those projected in the future, a similar analysis is performed using detrended RCP8.5 projections from the primary subset of models in Table 1. As shown in Fig. 10, the results are qualitatively similar to those of the pre-industrial control simulations, which further emphasizes the robustness of the results.
We find that low sea ice in winter is associated with warm winter surface temperatures over the Arctic and cold surface temperatures over Eurasia, consistent with previous studies using observations or reanalyses (Cohen et al., 2013; Mori et al., 2014; Kug et al., 2015). The Eurasian cooling is dynamically related to a strengthened Siberian high, again consistent with past work (Mori et al., 2014). Crucially however, we show that both the strengthened Siberian high and the Eurasian cooling are present several months before the low sea ice. In contrast, we find no evidence of a strengthened Siberian high or Eurasian cooling in the months following low winter sea ice. This suggests that the Eurasian cooling is driven by atmospheric circulation anomalies that precede and may contribute to low sea ice, but is not directly driven by low sea ice. This supports the conclusions of (Sato et al., 2014), (Sorokina et al., 2016), (Sun et al., 2016) and (McCusker et al., 2016), but is contrary to other studies that proposed a causal relationship between low sea ice and Eurasian cooling (Honda et al., 2009; Petoukhov and Semenov, 2010; Mori et al., 2014; Kug et al., 2015).
Finally, we examine relationships between regional sea ice anomalies and polar cap height. Similar to (Sun et al., 2015), we find that low Atlantic sector sea ice, specifically in the Barents-Kara Sea, is correlated with a weakened stratospheric polar vortex; and low Pacific sector sea ice, specifically in the Sea of Okhotsk, is correlated with a strengthened polar vortex. In both cases, the polar cap height anomalies precede low sea ice by several months and are associated with meridional heat flux anomalies that also precede the low sea ice. Thus, our analyses suggest that modified meridional eddy heat flux could contribute simultaneously to both a perturbed polar vortex and low sea ice.
